Return to README.txt CVS log | Up to [ELWIX - Embedded LightWeight unIX -] / embedaddon / sqlite3 / ext / async |
1.1 ! misho 1: ! 2: Normally, when SQLite writes to a database file, it waits until the write ! 3: operation is finished before returning control to the calling application. ! 4: Since writing to the file-system is usually very slow compared with CPU ! 5: bound operations, this can be a performance bottleneck. This directory ! 6: contains an extension that causes SQLite to perform all write requests ! 7: using a separate thread running in the background. Although this does not ! 8: reduce the overall system resources (CPU, disk bandwidth etc.) at all, it ! 9: allows SQLite to return control to the caller quickly even when writing to ! 10: the database, eliminating the bottleneck. ! 11: ! 12: 1. Functionality ! 13: ! 14: 1.1 How it Works ! 15: 1.2 Limitations ! 16: 1.3 Locking and Concurrency ! 17: ! 18: 2. Compilation and Usage ! 19: ! 20: 3. Porting ! 21: ! 22: ! 23: ! 24: 1. FUNCTIONALITY ! 25: ! 26: With asynchronous I/O, write requests are handled by a separate thread ! 27: running in the background. This means that the thread that initiates ! 28: a database write does not have to wait for (sometimes slow) disk I/O ! 29: to occur. The write seems to happen very quickly, though in reality ! 30: it is happening at its usual slow pace in the background. ! 31: ! 32: Asynchronous I/O appears to give better responsiveness, but at a price. ! 33: You lose the Durable property. With the default I/O backend of SQLite, ! 34: once a write completes, you know that the information you wrote is ! 35: safely on disk. With the asynchronous I/O, this is not the case. If ! 36: your program crashes or if a power loss occurs after the database ! 37: write but before the asynchronous write thread has completed, then the ! 38: database change might never make it to disk and the next user of the ! 39: database might not see your change. ! 40: ! 41: You lose Durability with asynchronous I/O, but you still retain the ! 42: other parts of ACID: Atomic, Consistent, and Isolated. Many ! 43: appliations get along fine without the Durablity. ! 44: ! 45: 1.1 How it Works ! 46: ! 47: Asynchronous I/O works by creating a special SQLite "vfs" structure ! 48: and registering it with sqlite3_vfs_register(). When files opened via ! 49: this vfs are written to (using the vfs xWrite() method), the data is not ! 50: written directly to disk, but is placed in the "write-queue" to be ! 51: handled by the background thread. ! 52: ! 53: When files opened with the asynchronous vfs are read from ! 54: (using the vfs xRead() method), the data is read from the file on ! 55: disk and the write-queue, so that from the point of view of ! 56: the vfs reader the xWrite() appears to have already completed. ! 57: ! 58: The special vfs is registered (and unregistered) by calls to the ! 59: API functions sqlite3async_initialize() and sqlite3async_shutdown(). ! 60: See section "Compilation and Usage" below for details. ! 61: ! 62: 1.2 Limitations ! 63: ! 64: In order to gain experience with the main ideas surrounding asynchronous ! 65: IO, this implementation is deliberately kept simple. Additional ! 66: capabilities may be added in the future. ! 67: ! 68: For example, as currently implemented, if writes are happening at a ! 69: steady stream that exceeds the I/O capability of the background writer ! 70: thread, the queue of pending write operations will grow without bound. ! 71: If this goes on for long enough, the host system could run out of memory. ! 72: A more sophisticated module could to keep track of the quantity of ! 73: pending writes and stop accepting new write requests when the queue of ! 74: pending writes grows too large. ! 75: ! 76: 1.3 Locking and Concurrency ! 77: ! 78: Multiple connections from within a single process that use this ! 79: implementation of asynchronous IO may access a single database ! 80: file concurrently. From the point of view of the user, if all ! 81: connections are from within a single process, there is no difference ! 82: between the concurrency offered by "normal" SQLite and SQLite ! 83: using the asynchronous backend. ! 84: ! 85: If file-locking is enabled (it is enabled by default), then connections ! 86: from multiple processes may also read and write the database file. ! 87: However concurrency is reduced as follows: ! 88: ! 89: * When a connection using asynchronous IO begins a database ! 90: transaction, the database is locked immediately. However the ! 91: lock is not released until after all relevant operations ! 92: in the write-queue have been flushed to disk. This means ! 93: (for example) that the database may remain locked for some ! 94: time after a "COMMIT" or "ROLLBACK" is issued. ! 95: ! 96: * If an application using asynchronous IO executes transactions ! 97: in quick succession, other database users may be effectively ! 98: locked out of the database. This is because when a BEGIN ! 99: is executed, a database lock is established immediately. But ! 100: when the corresponding COMMIT or ROLLBACK occurs, the lock ! 101: is not released until the relevant part of the write-queue ! 102: has been flushed through. As a result, if a COMMIT is followed ! 103: by a BEGIN before the write-queue is flushed through, the database ! 104: is never unlocked,preventing other processes from accessing ! 105: the database. ! 106: ! 107: File-locking may be disabled at runtime using the sqlite3async_control() ! 108: API (see below). This may improve performance when an NFS or other ! 109: network file-system, as the synchronous round-trips to the server be ! 110: required to establish file locks are avoided. However, if multiple ! 111: connections attempt to access the same database file when file-locking ! 112: is disabled, application crashes and database corruption is a likely ! 113: outcome. ! 114: ! 115: ! 116: 2. COMPILATION AND USAGE ! 117: ! 118: The asynchronous IO extension consists of a single file of C code ! 119: (sqlite3async.c), and a header file (sqlite3async.h) that defines the ! 120: C API used by applications to activate and control the modules ! 121: functionality. ! 122: ! 123: To use the asynchronous IO extension, compile sqlite3async.c as ! 124: part of the application that uses SQLite. Then use the API defined ! 125: in sqlite3async.h to initialize and configure the module. ! 126: ! 127: The asynchronous IO VFS API is described in detail in comments in ! 128: sqlite3async.h. Using the API usually consists of the following steps: ! 129: ! 130: 1. Register the asynchronous IO VFS with SQLite by calling the ! 131: sqlite3async_initialize() function. ! 132: ! 133: 2. Create a background thread to perform write operations and call ! 134: sqlite3async_run(). ! 135: ! 136: 3. Use the normal SQLite API to read and write to databases via ! 137: the asynchronous IO VFS. ! 138: ! 139: Refer to sqlite3async.h for details. ! 140: ! 141: ! 142: 3. PORTING ! 143: ! 144: Currently the asynchronous IO extension is compatible with win32 systems ! 145: and systems that support the pthreads interface, including Mac OSX, Linux, ! 146: and other varieties of Unix. ! 147: ! 148: To port the asynchronous IO extension to another platform, the user must ! 149: implement mutex and condition variable primitives for the new platform. ! 150: Currently there is no externally available interface to allow this, but ! 151: modifying the code within sqlite3async.c to include the new platforms ! 152: concurrency primitives is relatively easy. Search within sqlite3async.c ! 153: for the comment string "PORTING FUNCTIONS" for details. Then implement ! 154: new versions of each of the following: ! 155: ! 156: static void async_mutex_enter(int eMutex); ! 157: static void async_mutex_leave(int eMutex); ! 158: static void async_cond_wait(int eCond, int eMutex); ! 159: static void async_cond_signal(int eCond); ! 160: static void async_sched_yield(void); ! 161: ! 162: The functionality required of each of the above functions is described ! 163: in comments in sqlite3async.c. ! 164: